[cache][multi-arch] use apple_common.multi_arch_split

[jerrymarino]

This PR fixes #184

Today, there's redundant builds of a swift_library with rules_ios framework
bundle and rules_apple bundles like a an app or test.

Given:

apple_framework(name = "fw", ..)
ios_application(name="app", deps=["fw"])

When running the build, there will be 2 swiftmodules built ( 1 per the app, and 1 per the framework )

bazel build app fw

Consequently, it builds 2x of all the transitive swift libraries and also
causes issues with cache hits.

The root cause is that multi platform support isn't implemented in rules apple.
To partition up actions, we give the deps the transition
apple_common.multi_arch_split. This implicates a different configuration

Note that there is more changes to handle ios_multi_cpus
end to end #186

Testing - it doesn't seem possible to test this with an analysis test
with skylib given the way Bazel transtively computes the output
directory, e.g. for a target_under_test. When #186 is done adding a
CLI test with ios_multi_cpus will assert that it can build in this
way.

[amberdixon]

The approach makes sense to me, but left an open question about the new test introduced.

[segiddins]

Testing - it doesn't seem possible to test this with an analysis test
with skylib given the way Bazel transtively computes the output
directory, e.g. for a target_under_test. When #186 is done adding a
CLI test with ios_multi_cpus will assert that it can build in this
way.

I think we can use analysis tests to force into whatever config we want, and take multiple targets as deps and check that the subdir of bazel-out that their outputs go into matches?

[jerrymarino]

I think we can use analysis tests to force into whatever config we want, and take multiple targets as deps and check that the subdir of bazel-out that their outputs go into matches?

Do you have any suggestion on how to use an analysis test to force a config? I couldn't figure out a way to do this without a decently complex aspect, which might be more trouble than its worth. Anyhow, there's a few other issues I've gotta work out with the CI on this still.

[segiddins]

@jerrymarino https://github.com/bazelbuild/bazel-skylib/blob/master/docs/unittest_doc.md#analysistest.make-config_settings should allow changing config settings used for analysis of the target under test

[segiddins]

@jerrymarino it looks like this no longer has anything to do with apple_common.multi_arch_split, did you force-push the wrong branch?

[jerrymarino]

@segiddins thanks for digging into the bit on analysis test config - this is great. Once I've fixed the misc edges on this I suspect I can come up with an aspect to surface out all swift_library outputs and make assertions on the output root in the test. Likely would be a followup to keep this one somewhat straight forward.